observability

Observability with Dapr

This quickstart explores the observability capabilities of Dapr. Observability includes metric collection, tracing, logging and health checks. In this quickstart you'll be enabling distributed tracing on an application without changing any application code or creating a dependency on any specific tracing system. Dapr works with the Zipkin (great for development) and the Open Telemetry (recommended for production) protocols for distributed traces.

In this quickstart you will:

• Deploy Zipkin and configure it as a tracing provider for Dapr in self hosted mode and in Kubernetes.
• Configure an application for tracing and then deploy it.
• Troubleshoot a performance issue.

Configure self hosted mode

For self hosted mode, first run dapr init. When you run dapr init:

1. The following YAML file is created by default in $HOME/.dapr/config.yaml (on Linux/Mac) or %USERPROFILE%\.dapr\config.yaml (on Windows) and it is referenced by default on dapr run calls unless otherwise overridden:

• config.yaml

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: daprConfig
  namespace: default
spec:
  tracing:
    samplingRate: "1"
    zipkin:
      endpointAddress: "http://localhost:9411/api/v2/spans"

2. The openzipkin/zipkin docker container is launched.

3. The applications launched with dapr run will by default reference the config file in $HOME/.dapr/config.yaml or %USERPROFILE%\.dapr\config.yaml and can be overridden with the Dapr CLI using the --config param. For example, the following command will launch the hello-world quickstart app using the default config.yaml:

4. Clone this repo using git clone [-b <dapr_version_tag>] https://github.com/dapr/quickstarts.git and go to the repo's directory via cd quickstarts/tutorials/observability.

cd ../hello-world/node && npm install && dapr run --app-id hello-tracing --app-port 3000 node app.js && cd ../../observability

5. Once the app is running, you can make a request, which will populate at least one trace:

dapr invoke --app-id hello-tracing --method neworder --data-file sample.json

Viewing Traces

Tracing is set up out of the box when running dapr init. To view traces, in your browser go to http://localhost:9411 and you will see the Zipkin UI.

Zipkin API

Zipkin also has an API available. See Zipkin API for more details.

To see traces collected through the API:

curl -s "http://localhost:9411/api/v2/traces?spanName=calllocal%2Fhello-tracing%2Fneworder" -H "accept:application/json" -o output.json && python3 -m json.tool output.json

You should see output like the following:

[
    [
        {
            "traceId": "4c480d57b0e6d96b7150b46d027c5904",
            "id": "90d58917274e180a",
            "kind": "CLIENT",
            "name": "calllocal/hello-tracing/neworder",
            "timestamp": 1613170216016911,
            "duration": 93671,
            "localEndpoint": {
                "serviceName": "hello-tracing",
                "ipv4": "127.0.0.1"
            },
            "tags": {
                "dapr.api": "POST /v1.0/invoke/hello-tracing/method/neworder",
                "dapr.protocol": "http",
                "dapr.status_code": "200",
                "net.peer.name": "hello-tracing",
                "opencensus.status_description": "OK",
                "rpc.service": "ServiceInvocation"
            }
        }
    ]
]

Cleanup

dapr stop --app-id hello-tracing

Configure Kubernetes

Prerequisites

This quickstart builds on the distributed calculator quickstart and requires Dapr to be installed on a Kubernetes cluster along with a state store. It is suggested to go through the distributed calculator quickstart before this one. If you have not done this then:

1. Clone this repo using git clone [-b <dapr_version_tag>] https://github.com/dapr/quickstarts.git and go to the directory via cd quickstarts/tutorials/obervability.
2. Install Dapr on Kubernetes.
3. Configure Redis as a state store for Dapr.
4. Review the host and password for Redis state store Component in ../distributed-calculator/deploy/redis.yaml.

Note: See https://github.com/dapr/quickstarts#supported-dapr-runtime-version for supported tags. Use git clone https://github.com/dapr/quickstarts.git when using the edge version of dapr runtime.

Configure Dapr tracing in the cluster

Review the Dapr configuration file ./deploy/appconfig.yaml below:

apiVersion: dapr.io/v1alpha1
kind: Configuration
metadata:
  name: appconfig
spec:
  tracing:
    samplingRate: "1"
    zipkin:
      endpointAddress: "http://zipkin.default.svc.cluster.local:9411/api/v2/spans"

• samplingRate is used to enable or disable the tracing. To disable the sampling rate , set samplingRate : "0" in the configuration. The valid range of samplingRate is between 0 and 1 inclusive. The sampling rate determines whether a trace span should be sampled or not based on value. samplingRate : "1" will always sample the traces. By default, the sampling rate is 1 in 10,000.
• zipkin.endpointAddress is used to specify the trace backend to receive trace using the Zipkin format. Any backend that understands the Zipkin trace format, like Zipkin, Jaeger, New Relic, etc can be used. In this sample, we use the address of the Zipkin server that we will deploy in the next step.

This configuration file enables Dapr tracing. Deploy the configuration by running:

kubectl apply -f ./deploy/appconfig.yaml

You can see that now a new Dapr configuration which enables tracing has been added. Run the command:

dapr configurations --kubernetes

You should see output that looks like this:

  NAME       TRACING-ENABLED  METRICS-ENABLED  AGE  CREATED
  appconfig  true             true             1h   2020-12-10 22:01.59

You can see that appconfig has TRACING-ENABLED set to true.

Deploy Zipkin to the cluster and set it as the tracing provider

In this quickstart Zipkin is used for tracing. Examine ./deploy/zipkin.yaml and see how it includes three sections:

1. A Deployment for Zipkin using the openzipkin/zipkin docker image.
2. A Service which will expose Zipkin internally as a ClusterIP in Kubernetes.

Deploy Zipkin to your cluster by running:

kubectl apply -f ./deploy/zipkin.yaml

Now that Zipkin is deployed, you can access the Zipkin UI by creating a tunnel to the internal Zipkin service you just created by running (with port 19411 chosen to avoid conflicts with Dapr CLI running Zipkin in self-hosted mode):

kubectl port-forward svc/zipkin 19411:9411

On your browser go to http://localhost:19411. You should be able to see the Zipkin dashboard.

Instrument the application for tracing and deploy it

To instrument a service for tracing with Dapr, no code changes are required, Dapr handles all of the tracing using the Dapr side-car. All that is needed is to add the Dapr annotation for the configuration you deployed earlier (which enables tracing) in the application deployment yaml along with the other Dapr annotations. The configuration annotation looks like this:

...
annotations:
...
    dapr.io/config: "appconfig"
...

For this quickstart, a configuration has already been enabled for every service in the distributed calculator app. You can find the annotation in each one of the calculator yaml files. For example review the yaml file for the calculator front end service here.

Note you did not introduce any dependency on Zipkin into the calculator app code or deployment yaml files. The Zipkin Dapr component is configured to read tracing events and write these to a tracing backend.

Now deploy the distributed calculator application to your cluster:

kubectl apply -f ../distributed-calculator/deploy

Kubernetes deployments are asyncronous. This means you'll need to wait for the deployment to complete before moving on to the next steps. You can do so with the following commands:

kubectl rollout status deploy/addapp

kubectl rollout status deploy/subtractapp

kubectl rollout status deploy/divideapp

kubectl rollout status deploy/multiplyapp

kubectl rollout status deploy/calculator-front-end

You can view the status of the running pods with:

kubectl get pods

Then, open the distributed calculator UI.

If this is the first time trying the distributed calculator, find more detailed instructions in the distributed-calculator tutorial.

Note: If the distributed calculator is already running on your cluster you will need to restart it for the tracing to take effect. You can do so by running:

kubectl rollout restart deployment addapp calculator-front-end divideapp multiplyapp subtractapp

Optional: if you don't have easy public browser access, you can always use port forwarding

kubectl port-forward service/calculator-front-end 8000:80

Discover and troubleshoot a performance issue using Zipkin

To show how observability can help discover and troubleshoot issues on a distributed application, you'll update one of the services in the calculator app. This updated version simulates a performance degradation in the multiply operation of the calculator that you can then investigate using the traces emitted by the Dapr sidecar. Run the following to apply a new version of the python-multiplier service:

kubectl apply -f ./deploy/python-multiplier.yaml

As above, you can wait for the asyncronous Kubernetes deployment with the following:

kubectl rollout status deploy/multiplyapp

Now go to the calculator UI and perform several calculations. Make sure to use all operands. For example, do the following:

9 + 3 * 2 / 4 - 1 =

Optional: You can also use the following curl commands to execute all operations:

curl -s http://localhost:8000/calculate/add -H Content-Type:application/json --data @operands.json

curl -s http://localhost:8000/calculate/subtract -H Content-Type:application/json --data @operands.json

curl -s http://localhost:8000/calculate/divide -H Content-Type:application/json --data @operands.json

curl -s http://localhost:8000/calculate/multiply -H Content-Type:application/json --data @operands.json

curl -s http://localhost:8000/persist -H Content-Type:application/json --data @persist.json

curl -s http://localhost:8000/state 

Now go to the Zipkin dashboard by running. (Note: if you are running Dapr locally, be sure to use a different local port for Zipkin):

kubectl port-forward svc/zipkin 19411:9411

And browsing to http://localhost:19411. Click the search button to view tracing coming from the application:

Dapr adds a HTTP/gRPC middleware to the Dapr sidecar. The middleware intercepts all Dapr and application traffic and automatically injects correlation IDs to trace events. You can see a lot of transactions are being captured including the regular health checks done by Kubernetes:

Now look for any performance issues by filtering on any requests that have take too long. You can use minDuration criteria to query for long requests only:

You can quickly see that the multiply method invocation is unusually slow (takes over 1 second). Since the problem may be either at the calculator-frontend service or the python-multiplier service you can dig further by clicking on the entry:

Now you can see which specific call was delayed via the data field (here it's the 12 * 2 operation) and confirm that it is the multiplier service which you updated that is causing the slowdown (You can find the code for the slow multiplier under the python directory).

Zipkin API

As before, you can also access traces through the Zipkin API. The following will give you the same traces as the UI search above:

curl -s http://localhost:19411/api/v2/traces?minDuration=250000 -H accept:application/json -o output.json && python3 -m json.tool output.json

You should get output like this:

[
    [
        {
            "duration": 1009084,
            "id": "ff0bf110ca88f770",
            "kind": "SERVER",
            "localEndpoint": {
                "ipv4": "10.244.4.225",
                "serviceName": "multiplyapp"
            },
            "name": "calllocal/multiplyapp/multiply",
            "parentId": "733a1812e839dcfb",
            "tags": {
                "dapr.api": "/dapr.proto.internals.v1.ServiceInvocation/CallLocal",
                "dapr.invoke_method": "multiply",
                "dapr.protocol": "grpc",
                "rpc.service": "ServiceInvocation"
            },
            "timestamp": 1613177766316506,
            "traceId": "926f1a5a6c63ae6f5167c29b3ddf4271"
        },
...

Clean up

1. To remove the distributed calculator application from your cluster run:

kubectl delete -f ../distributed-calculator/deploy

2. To remove the Zipkin installation and tracing configuration run:

kubectl delete -f deploy/zipkin.yaml

Additional Resources

• Learn more about observability.
• Learn more on how Dapr does distributed tracing.

Next steps

• Explore additional quickstarts

Troubleshooting

If you see an error with the following message:

time="2021-02-13T00:39:00.48769561Z" level=fatal msg="process component zipkin error: incorrect type exporters.zipkin" app_id=addapp instance=addapp-59f447d8b6-w48jt scope=dapr.runtime type=log ver=1.0.0-rc.4

It means there is an old exporter Component in your Kubernetes cluster. First, list Components:

kubectl get components

NAME         AGE
zipkin       71d
statestore   43m

Then, identify which Component is the exporter (usually named zipkin) and delete it:

kubectl delete component zipkin